Subscribe Bookmark RSS Feed

Meaning of the parameter estimates


Community Member


May 17, 2017

I am studying what discourages men and women from working in another country. I have my table and I can conclude that women are more discouraged by political factors than men with a significance level of 0,0017. But I am not entirely sure that I understand the "Estimate". What can I say about the -1,2699? 


The score is a ranking of how a factor discourage desire to live and work in that country. From 1 - I does not discourage me, to 5 - It will very much discourage me. 

gender 2.PNG






Jul 7, 2014

It looks you fit a binary logistic regression with several continuous factors. So, each estimates measures the log odds ratio (aks logit) of a factor. For example, for Politic, it shows that a one unit increase in Politic leads to a decrease in log odds ratio by 1.2699. A more widely used measure is odds ratio, which can be requested in JMP


Some additional online resources on logistic regression parameters:





Jun 23, 2011

You using the Nominal Logistic Fit platform. You say that the response has five levels. If that is true, then you would have a different report. Does the column with the response really have five levels (1-5)? Is it using character or numeric data type? Is it using the ordinal modeling type? Such a response is ordinal in nature so you can take advantage of the extra information about rankings as well as differences. This arrangement should automatically lead to the Ordinal Logistic Fit platform (starting with Analyze > Fit Model).

The parameter estimate is coefficient of your linear predictor. So it represents the change in the response if you have a particular level of a categorical predictor or a change of 1 unit for a continuous predictor. It means the same thing as in a multiple regression analysis with a continuous response.

The difference here is the response. It is continuous through a transformation: the logit. You are fitting the linear predictor to the log( odds ). The ordinal response uses the cumulative logit, so you are fitting the linear predictor to the log( Pr( level i or below ) / Pr( level i+1 or higher ). In your case you will have four logits (the number of logits is always one less than the number of response levels). In effect, you have four sets of parameter estimates but they use a common intercept in the case of an ordinal response.

You can observe the explicit form of the ordinal logistic regression model by selecting Save Columns > Prediction Formula and then examining the data table. You will have one column for the linear predictor and then the back transformation for the probability of each level. Finally, you will have the prediction, which is the level with the highest probability.

Learn it once, use it forever!